def main():
# Retrieve data
file_path = "data/interim/train_interim.csv"
housing_prices = pd.read_csv(file_path)
# Seperating predictors and target
input_feats, output_feats = preprocessing.make_dataset(
housing_prices, "SalePrice")
# Subsetting columns of interest
feature_names = [
"LotArea",
"YearBuilt",
"1stFlrSF",
"2ndFlrSF",
"FullBath",
"BedroomAbvGr",
"TotRmsAbvGrd",
"HouseStyle",
]
features = input_feats[feature_names]
# Data processing
preprocess_pipeline = preprocessing.preprocess_pipeline(features)
# Generating pipeline for model
model = make_pipeline(preprocess_pipeline, LinearRegression())
# Train the model
model, predictions, actual = train_model.train_model(
features, output_feats, model)
# Evaluating the model
train_model.evaluate_model(predictions, actual, model)
def parameterized_test(self, model, mode):
# given:
data_dir = "test-data"
interim_dir = self.test_dir + "/interim"
processed_dir = self.test_dir + "/processed"
model_dir = self.test_dir + "/model"
model_path = model_dir + ("" if mode == "full" else "_" +
mode) + "/0001.txt"
submission_dir = self.test_dir + "/submissions"
submission_path = submission_dir + "/submission.csv"
# data preparation
# when:
make_dataset(data_dir, interim_dir)
# then:
self.assertTrue(os.path.exists(interim_dir + "/test_data.pkl"))
self.assertTrue(os.path.exists(interim_dir + "/test_data.pkl"))
# feature engineering
# when:
build_features(data_dir, processed_dir)
# then:
self.assertTrue(os.path.exists(processed_dir + "/test_data.pkl"))
self.assertTrue(os.path.exists(processed_dir + "/test_data.pkl"))
# model training
# when:
train_model(model, mode, processed_dir, model_dir)
# then:
self.assertTrue(os.path.exists(model_path))
# model prediction
# when:
predict_model(processed_dir, model, model_path, submission_path)
# then:
self.assertTrue(os.path.exists(submission_path))
def main_test():
start_time = time.time()
train = read_train_data(nrows=None)
test = read_test_data()
train, test = process_data(train, test)
X = train.drop(['ID_code', 'target'], axis=1)
y = train['target']
X_test = test.drop(['ID_code'], axis=1)
oof, predictions, scores, feature_importance = train_model(
X, X_test, y, params, plot_feature_importance=True)
str_metric_score = metric + '_0' + str(
int(scores['auc_score'].iloc[0] * 10000))
# submit(test, predictions, str_metric_score)
comment = 'starter removed statistics feature, remove also 0 score, bagging_fraction1'
# storage_src(str_metric_score, scores, feature_importance, comment)
elapsed_time = time.time() - start_time
print(elapsed_time)
def main():
# Retrieve data
file_path = "data/interim/train_interim.csv"
housing_prices = pd.read_csv(file_path)
# Seperating predictors and target
input_feats, output_feats = preprocessing.make_dataset(housing_prices, "SalePrice")
# Subsetting columns of interest
feature_names = [
"LotArea",
"YearBuilt",
"1stFlrSF",
"2ndFlrSF",
"FullBath",
"BedroomAbvGr",
"TotRmsAbvGrd",
"HouseStyle",
]
features = input_feats[feature_names]
# Data processing
preprocess_pipeline = preprocessing.preprocess_pipeline(features)
# Generating pipeline for model
pipeline = make_pipeline(preprocess_pipeline, KNeighborsRegressor())
# Defining a params for grid-search
params = {
"kneighborsregressor__n_neighbors": range(2, 21),
"kneighborsregressor__weights": ["uniform", "distance"],
}
model = GridSearchCV(pipeline, params, cv=10, scoring="neg_mean_squared_error")
# Train the model
model, predictions, actual = train_model.train_model(features, output_feats, model)
# check the best parameters that was chosen
print(f"Best parameters chosen: {model.best_params_}")
# Evaluating the model
train_model.evaluate_model(predictions, actual, model)
def main_submit():
start_time = time.time()
train = read_train_data(nrows=None)
test = read_test_data()
train, test = process_data(train, test)
X = train.drop(['ID_code', 'target'], axis=1)
y = train['target']
X_test = test.drop(['ID_code'], axis=1)
oof, predictions, scores, feature_importance = train_model(
X,
X_test,
y,
params,
n_fold=10,
plot_feature_importance=True,
model_type='lgb_sklearn')
str_metric_score = metric + '_0' + str(
int(scores['auc_score'].iloc[0] * 10000))
submit(test, predictions, str_metric_score)
comment = 'add 5 max min feature before standard scale'
storage_src(str_metric_score, scores, feature_importance, comment)
elapsed_time = time.time() - start_time
print(elapsed_time)
import numpy as np
import pandas as pd
from src.models.train_model import train_model
dic = train_model()
model = dic['model']
vect = dic['vect']
def predict_class(row):
row['class_cat'] = model.predict(vect.transform([row['feat_name']]))[0]
probabilities = list(model.predict_proba(
vect.transform([row['feat_name']])))[0]
row['probabilities'] = round(max(probabilities), 2)
if row['class_cat'] == 0:
row['class'] = 'Das'
if row['class_cat'] == 1:
row['class'] = 'Der'
if row['class_cat'] == 2:
row['class'] = 'Die'
return row
def gen_df_results():
feat_value = model.coef_[0]
order_of_importance = (-feat_value).argsort()
feat_names = np.array(vect.get_feature_names())
dic_results = {'feat_name': feat_names[order_of_importance],
def run(all_code_types,
d_embedding,
embedding_dropout_p,
min_count,
batch_size,
verbose,
epochs,
lr,
wd,
logsig,
sig_depth,
run_name,
patience,
add_time,
leadlag,
t_scale,
t_max,
use_timestamps,
feedforward_num_layers,
feedforward_hidden_dims,
feedforward_activations,
feedforward_dropout,
training_proportion=1,
testing_subsample_size=None,
split_paths=False,
tensorboard_log=False,
evaluate_on_test=True):
"""Run the experiment for either cross validation or testing"""
dataset, dataset_test, vocab = generate_ml_data(
all_code_types,
min_count,
batch_size,
verbose=verbose,
allen_mode=True,
dataset_path=None,
training_proportion=training_proportion,
testing_subsample_size=testing_subsample_size,
split_paths=split_paths)
logger.info("Using k-fold cross validation")
# Allen kfold
metrics_by_fold = []
cross_validator = StratifiedKFold(n_splits=K_FOLDS, shuffle=True)
n_splits = cross_validator.get_n_splits(dataset)
for fold_index, (train_indices, validation_indices) in enumerate(
cross_validator(dataset)):
logger.info(f"Fold {fold_index}/{n_splits - 1}")
train_dataset = Subset(
dataset,
train_indices,
)
validation_dataset = Subset(dataset, validation_indices)
train_loader = DataLoader(dataset=train_dataset,
batch_size=batch_size,
shuffle=True)
validation_loader = DataLoader(dataset=validation_dataset,
batch_size=batch_size,
shuffle=True)
if tensorboard_log or evaluate_on_test:
serialization_dir = os.path.join(TENSORBOARD_DIR, run_name,
str(uuid.uuid4()),
str(fold_index))
else:
serialization_dir = None
model = init_sig(vocab, d_embedding, embedding_dropout_p, sig_depth,
logsig, all_code_types, feedforward_num_layers,
feedforward_hidden_dims, feedforward_activations,
feedforward_dropout, leadlag, add_time, t_max,
t_scale, use_timestamps, split_paths)
if torch.cuda.is_available():
cuda_device = 0
model = model.cuda(cuda_device)
logger.info('USING CUDA GPU')
else:
cuda_device = -1
fold_metrics, model = train_model(model, lr, wd, train_loader,
validation_loader, patience, epochs,
cuda_device, serialization_dir)
if serialization_dir is not None:
ex.add_artifact(
os.path.join(serialization_dir,
'best.th')) # Add file location to sacred log
metrics_by_fold.append(fold_metrics)
if evaluate_on_test:
if serialization_dir is None:
raise Exception(
'serialization_dir needed to load best model from validation'
)
test_dataloader = DataLoader(dataset=dataset_test,
batch_size=batch_size,
shuffle=True) # Held out test data
metrics = evaluate(model, test_dataloader, cuda_device)
return metrics
torch.cuda.empty_cache()
metrics = reformat_metrics(metrics_by_fold, ex)
return metrics
def main():
# Sidebar section:
page_selection = st.sidebar.radio("Select a market:",
["Nikkey", "Bovespa"])
dct_market = {
"Nikkey": {
"country": "Japan",
"continent": "Asia",
"index_name": "^N225"
},
"Bovespa": {
"country": "Brazil",
"continent": "America",
"index_name": "^BVSP"
}
}
st.markdown(f"# {page_selection}")
end_date = date.today()
start_date = end_date - timedelta(days=3150)
# start_date = datetime.strptime('2004-11-02', '%Y-%m-%d')
# end_date = datetime.strptime('2008-11-28', '%Y-%m-%d')
start_date = st.sidebar.date_input('Start date', start_date)
end_date = st.sidebar.date_input('End date', end_date)
df = yf.download(dct_market[page_selection]["index_name"],
start=start_date,
end=end_date)
df["rt"] = (np.log(df["Close"]) -
np.log(df["Close"].shift(periods=1))) * 100
df = create_shifted_rt(df, [1, 5, 37])
df_clustered = uniform_clustering(
df[["Close", "rt", "rt-1", "rt-5", "rt-37"]],
["rt", "rt-1", "rt-5", "rt-37"])
df_clustered.dropna(how="any", axis=0, inplace=True)
lst_relations = [('cluster_rt-37', 'cluster_rt'),
('cluster_rt-5', 'cluster_rt'),
('cluster_rt-1', 'cluster_rt')]
df_clustered = df_clustered[[
"rt", "cluster_rt-37", "cluster_rt-5", "cluster_rt-1", "cluster_rt"
]]
predict_n_days = 20
model = train_model(df_clustered.iloc[:-predict_n_days], lst_relations)
evidence = {
'cluster_rt-37': df_clustered.iloc[-37]['cluster_rt'],
'cluster_rt-5': df_clustered.iloc[-5]['cluster_rt'],
'cluster_rt-1': df_clustered.iloc[-1]['cluster_rt']
}
predict = predict_model(model, evidence=evidence)
st.text(f"Previsão para amanhã: {predict[0]}")
resultado = {}
for i in np.arange(1, predict_n_days + 1):
evidence = {
'cluster_rt-37': df_clustered.iloc[-37 - i]['cluster_rt'],
'cluster_rt-5': df_clustered.iloc[-5 - i]['cluster_rt'],
'cluster_rt-1': df_clustered.iloc[-1 - i]['cluster_rt']
}
predict = predict_model(model, evidence=evidence)
resultado[i] = [
predict[0]['cluster_rt'], df_clustered.iloc[i]['cluster_rt'],
df_clustered.iloc[i]['rt']
]
resultado = pd.DataFrame.from_dict(resultado, orient='index')
resultado.rename(columns={0: 'Previsão', 1: 'Real', 2: 'rt'}, inplace=True)
rt_mean = round(
resultado.groupby(by=["Real"]).agg(
{"rt": ["min", "max", "count", "mean"]}), 2)[("rt", "mean")]
if page_selection == "Nikkey":
conditions = [
resultado["Previsão"] == 1.0, resultado["Previsão"] == 2.0,
resultado["Previsão"] == 3.0, resultado["Previsão"] == 4.0,
resultado["Previsão"] == 5.0, resultado["Previsão"] == 6.0
]
elif page_selection == "Bovespa":
conditions = [
resultado["Previsão"] == 1.0, resultado["Previsão"] == 2.0,
resultado["Previsão"] == 3.0, resultado["Previsão"] == 4.0
]
choices = rt_mean.tolist()
resultado["rt_predict"] = np.select(conditions, choices, default=np.nan)
resultado = resultado[::-1]
resultado["rt_predict_acumulado"] = resultado["rt_predict"].cumsum()
resultado["rt_acumulado"] = resultado["rt"].cumsum()
st.dataframe(resultado)
rmse_uniform = mean_squared_error(resultado["rt"],
resultado["rt_predict"],
squared=False)
acuracia = accuracy_score(resultado["Real"],
resultado["Previsão"],
normalize=True)
st.text(f"Acurácia: {round(acuracia*100, 2)}%")
st.text(f"RMSE: {round(rmse_uniform, 2)}%")
# fig = plt.figure(figsize=(20, 4))
# ax = fig.add_subplot(111)
# ax.plot(df['Close'], label=dct_market[page_selection]["index_name"])
# date_min = df.index.min()
# date_max = df.index.max()
# ax.xaxis.set_major_locator(plt.MaxNLocator(30))
# ax.set_xlim(left=date_min, right=date_max)
# ax.legend(loc='lower left', frameon=False)
# plt.xticks(rotation=90)
# st.pyplot(fig)
st.line_chart(df[['Close']])
st.line_chart(df["rt"])
"""Use prediction model and evaluate it"""
from os import path
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score
from src.models import train_model
from src.visualizations import Visualize
base_path ='/home/chpatola/Desktop/Skola/Python/cookie_nlp/'
#1. Test model and look into results
test_X, test_y, bag, tf_idf = train_model.train_model(base_path)
print("Mean accurancy in validation: {:.2f} %".format(100*bag.best_score_))
predictions = bag.predict(test_X)
print("Predictions:\n {} \nTruth:\n {}".format(predictions[0:3], test_y[0:3]))
print(test_X[0:3])
#2. Save classification report and confusion matrix to file
classi_rep = Visualize._plot_classification_report(test_y, predictions)
classi_rep.savefig(
path.join(base_path,'reports/figures/classificationReport.png'),
bbox_inches='tight')
parties = test_y.sort_values().unique()
Visualize.cm_analysis(test_y,
predictions,
path.join(base_path,'reports/figures/confusion_matrix.png'),
labels=parties
)
#3. Print results
print(confusion_matrix(test_y, predictions))
print(classification_report(test_y, predictions))
